Grafana Billing Metrics Skill

Query key billing metrics from Prometheus and Loki through Grafana's data source proxy API.

Quick Start

Query both staging and prod (default)

uv run .claude/skills/grafana-billing/scripts/billing_metrics.py

Query specific environment

uv run .claude/skills/grafana-billing/scripts/billing_metrics.py --env staging uv run .claude/skills/grafana-billing/scripts/billing_metrics.py --env prod

JSON output for automation

uv run .claude/skills/grafana-billing/scripts/billing_metrics.py --json

Filter to specific service

uv run .claude/skills/grafana-billing/scripts/billing_metrics.py --service prometheus uv run .claude/skills/grafana-billing/scripts/billing_metrics.py --service loki

Environment Variables Required

• GRAFANA_STAGING_API_KEY

• API key for staging Grafana workspace

• GRAFANA_PROD_API_KEY

• API key for prod Grafana workspace

Key Metrics Captured

Prometheus

Metric Description

Active Time Series Current count of active series (billing dimension)

Samples/sec Ingestion rate (DPM = samples/sec * 60)

TSDB Storage On-disk storage bytes

Top Cardinality Top 10 metrics by series count

Loki

Metric Description

Ingestion Rate GB/day being ingested

Total Bytes Cumulative bytes received

Active Streams Number of active log streams

Memory Chunks Chunks held in memory

When to Use

Use this skill when the user asks about:

• Observability billing or costs

• Active time series counts

• Prometheus cardinality analysis

• Loki ingestion rates

• Storage usage for metrics or logs

• Comparing staging vs production usage

Instructions for Claude

• Run the billing metrics script to gather current data

• Present the results in a clear, formatted way

• Highlight any concerning metrics (high cardinality, rapid growth)

• Compare staging vs prod if both are queried

• Suggest cost optimization if metrics are unusually high

Critical Rules

• Always check that API keys are set before running

• Use --json flag when you need to process the output programmatically

• Default to querying both environments for comparison

• Handle errors gracefully - missing data sources should not crash the script